The present invention is a rotating priority queue manager. A rotating priority queue manager in accordance with the present invention may include a plurality of source data channels, a corresponding plurality of processing resources, and an arbitrating interface directing the flow of data from the source channels to the processing resources where the data must flow over a shared data path.
The plurality of processing resources may comprise any system of parallel processors where the servicing of input data must be carried out in a manner where there the maximum latency for processing a given data channel is determinable, the arbitration between channels is equal, no input channel may prevent another channel from being serviced, and lower priority processing resources are not prohibited from receiving input data if higher priority processing resources are not currently available or if higher priority data is not currently available.
|
1. A system for arbitrating access to a shared data path comprising:
a plurality of n data sources;
a plurality of n processing resources;
a shared data path linking the plurality of data source channels and the plurality of processing resource channels;
a rotating priority queue;
switching circuitry; and
interconnections linking the data sources, rotating priority queue, switching circuitry, shared data path and processing resources,
wherein each data source corresponds to one processing resource,
wherein each data source/processing resource pair has a common channel number designation,
wherein the rotating priority queue controls access to the shared data path by the data source/processing resource pairs, and
wherein the switching circuitry routes one of the n data sources across the shared data path to its corresponding processing resource channel.
16. A method for arbitrating access to a shared data path, the method comprising:
receiving signals from a plurality of n data source channels,
receiving signals from a plurality of n processing resources,
waiting until signals from one or more data source channels and one or more processing resource channels are received, at least one data source channel and at least one processing resource channel having a common channel number;
comparing the at least one common channel number to contents of a plurality of n n-bit priority registers, each priority register having an inherent priority in relation to the other n−1 priority registers,
granting access to the shared data path by a data source/processing resource channel pair;
transitioning the priority registers to a next state; and
deasserting the signals from the data source channel and the processing resource channel of the data source/processing resource channel pair which is permitted to access the shared data path.
2. The system of
wherein the maximum latency for access to the shared data path by any of the n data source/processing resource pairs is determinable.
3. The system of
wherein the rotating priority queue comprises:
a plurality of n priority registers,
wherein the priority queue receives input signals from the data sources and processing resources,
wherein the priority queue transmits an output signal to the switching circuitry, and
wherein each of the priority registers has an inherent priority with respect to the remaining n−1 priority registers.
4. The system of
wherein the plurality of data source channels and processing resource channels possess no inherent priority.
5. The system of
wherein the maximum latency for access to the shared data path by any of the n data source/processing resource pairs is only function of the number of channels.
6. The system of
wherein each of the priority registers is an n-bit register,
wherein each of the priority registers maintains a binary value such that only one of the n-bits is asserted at any time, the least significant bit having a zero-position and the nth bit having an n−1-position,
wherein no two priority registers have the same bit of their n-bit binary value asserted, and
wherein the priority of a data source/processing resource pair at a given time is equal to that of the priority register whose asserted-bit-position corresponds to the common channel number of the channel pair.
7. The system of
wherein the input signal from a data source channel is asserted when the data source channel currently has data pending for processing, and
wherein the input signal from a processing resource channel is asserted when the processing resource channel is currently available for accepting and processing data.
8. The system of
wherein the maximum latency for access to the shared data path by any of the n data source/processing resource pairs is n clock cycles.
9. The system of
wherein the next state of the priority registers is based on the current state of the registers.
10. The system of
wherein the next state of the priority registers is based on the input signals from the data source channels and the input signals from the processing resource channels.
11. The system of
wherein the next state of each priority register is calculated by rotating the binary value contained therein.
12. The system of
wherein the degree of rotation is based on the priority of the data source/processing resource channel pair which was most recently permitted to access the shared data path.
13. The system of
wherein, the asserted bit of each priority register is rotated i positions when the ith-priority data source/processing resource channel pair is permitted access to the shared data path.
17. The method of
wherein the maximum latency for access to the shared data path by any of the n data source/processing resource channel pairs is determinable.
18. The method of
wherein the signals from the plurality of n data source channels indicate that at least one data source has data available for processing, and
wherein the signals from the plurality of n processing resource channels indicate that at least one processing resource is ready to process input data.
19. The method of
wherein the step of comparing further comprises:
comparing the at least one common channel number to the contents of each of the plurality of n priority registers until at least one priority register's contents correspond to the at least one common channel number, the at least one priority register having an ith priority in relation to the other n−1 priority registers, i being from 1 to n.
20. The method of
wherein the switching circuitry is configured such that the data source/processing resource channel pair having the common channel number corresponding to the current highest-priority priority register is permitted to access the shared data path.
21. The method of
wherein the next state of the priority registers is based on the current state of the registers.
22. The method of
wherein the maximum latency for access to the shared data path by any of the n data source/processing resource channel pairs is a function only of the number of channels n.
23. The method of
wherein each of the n priority registers maintains an n-bit binary value such that only one of the n-bits is asserted at a given time, the least significant bit having the zero-position and the nth bit having the n−1-position,
wherein no two priority registers of the n priority registers have the same bit of their n-bit binary value asserted, and
wherein the at least one common channel number corresponds to the contents of a given priority register if the given priority register contains a value having an asserted-bit-position which equals the common channel number.
24. The method of
wherein the next state of the priority registers is based on the input signals from the data source channels and the input signals from the processing resource channels.
25. The method of
wherein the maximum latency for access to the shared data path by any of the data source/processing resource channel pairs is n clock cycles.
26. The method of
wherein the next state of the priority registers is determined by rotating the binary numbers contained in the registers.
27. The method of
wherein the degree of rotation is based on the priority of the data source/processing resource channel pair which was most recently permitted to access the shared data path.
28. The method of
wherein, the asserted bit of each priority register is rotated i positions when the ith-priority data source/processing resource channel pair is permitted access to the shared data path.
|
|||||||||||||||||||||||||||
This invention relates generally to processing systems and more particularly to systems where multiple data inputs are concurrently feeding multiple processing resources (on a one-to-one basis) where the data must flow over a shared data path.
Today's data-intensive audio, video and communications applications require high-bandwidth systems capable of real-time processing. As such, a primary goal in the design of modern processing units is maximizing throughput.
One of the most prevalent mechanisms for enhancing throughput in computing machines is the use of parallel processing. Parallel processing is the simultaneous execution of a set of tasks (split up and specially adapted) on multiple processors on multiple data segments in order to obtain results faster.
Systems having multiple interfaces linking the data sources and the processing resources are generally disfavored in circuit design as they can result in routing and pin congestion issues in application specific integrated circuits (ASIC). Routing a high number of signals requires physical space which is always at a premium in today's ever shrinking processing units. Likewise, for a given physical package, a predetermined pin count may already exist.
Instead, single-interface systems are commonly implemented. These single-interface systems utilize a data routing arbitration mechanism to establish a channel hierarchy to facilitate the transfer of data from multiple sources to multiple processors across a shared data path. Such systems, while more complex, require less board space and are more flexible in adapting to the pin requirements of existing physical packages.
However, the use of single-interface systems may also operate as a bottleneck to the required rapid transmission of data. In order to achieve the benefits of parallel computing, it is imperative that data routing to the respective processors occur in an efficient manner. The efficiency of the data arbitration algorithm is a primary factor in determining the overall throughput of a given parallel system. If only a few data and processing channels are allowed to dominate the interface, there is little or no advantage to incorporating a parallel system. Additionally, processors may require varying amounts of time to process certain data elements. As such, data channels must be routed to the various processors in a judicious manner such that no processing channel is starved if its corresponding data channel has pending data and, similarly, no pending data is blocked if its corresponding processing channel is requesting data.
An example of an application in which multiple parallel requesters are accessing a single resource, where a fair (equal priority) arbitration scheme is desired is cryptography. Cryptography systems are often utilized to prevent the interception of data when transmitted over public networks. The systems scramble data to prevent unauthorized access. Data segments may be processed by one of a plurality of encoders and then again by one of a plurality of decoders. Due to the nature of encrypt/decrypt systems, a given block of data may take more or less time to process than another block of equal size. As such, a cryptographic system must be able to route any data source channel to its corresponding processing channel at any time in order to maximize throughput.
However, the task of coordinating which data segments are to be routed to a given processor can be a complex task. For this reason, many priority schemes have been developed and implemented which govern the direction of data to associated parallel processors.
Simple static arbitration schemes have been used to direct data source channels having varying degrees of priority. In such schemes, when an interface receives an indication that data is pending on a high-priority channel, the transmission of data on lower-priority channels to the processors is halted until the higher-priority data is processed. However, this scheme is ineffective in applications where there is no inherent priority between the source channels (i.e. fair arbitration).
Additionally, dynamic rotating arbitration schemes have also been utilized in source data routing. These schemes are generally capable of varying their channel priority hierarchy. However, such systems, while more robust in their ability to handle priority-associated data sources, still suffer from the same channel-priority dependence as the static schemes. Their priority decisions, while fluid in nature, remain dependent on a predetermined dynamic rotation. Such schemes are unable to make decisions based on data source availability and processing resource readiness.
Therefore, it would be desirable to provide a system and a method for arbitrating between multiple data source/processing resource channels where the channels have equal priority, the arbitration has a determinable maximum latency, and no processing channel is starved or blocked.
Further, it would be desirable to provide a system and a method for arbitrating between multiple data source/processing resource channels where the availability of the data source and processing resources dictate the function of the arbitration.
Accordingly, the present invention is directed to a rotating priority queue manager. In general, the invention applies to systems wherein multiple inputs concurrently feed multiple processing resources which are organized to process data in a parallel arrangement where the data must flow over a shared data path.
In an embodiment of the invention, a priority queue manager may include a plurality of n data source channels, a corresponding plurality of n processing resource channels, and an arbitrating interface directing the flow of data from the source channels to the processing resources. Each corresponding data source/processing resource channel pair has an assigned common channel number.
The plurality of processing resources may comprise any system of parallel processors where the servicing of input data must be carried out in a manner where the maximum latency for processing a given data channel is determinable, the arbitration between channels is equal, no input channel may prevent another channel from being serviced, and lower priority processing resources are not prohibited from receiving input data if higher priority processing resources are not currently available or if higher priority data is not currently available. In a particular embodiment, the plurality of processors comprises multiple encrypt/decrypt engines for the encryption or decryption of data.
The arbitration interface comprises a Rotating Priority Queue (RPQ). The RPQ is comprised of a series of logic gates and a plurality of n-bit priority registers. Each of the n priority registers has a designated priority from highest (PR1) to lowest (PRn). The registers may be initialized with n-bit binary numbers as follows: PR1=20, PR2=21, PR3=22, . . . PRn-1=2n-2, PRn=2n-1, or generally PRi=2i-1, where i=1 to n.
The contents of these registers determine which of the n data source/processing resource channel pairs is assigned a given priority at a given time. Each of the priority registers maintains a value where only one of the n-bits is asserted as ‘1’ at any given time. The position of the asserted bit is defined such that the least significant bit is bit zero and the most significant bit is bit n−1. When the position of an asserted bit corresponds to the common channel number of a given data source/processing resource channel pair, that channel pair is assigned a priority equal to that of the associated priority register.
For example, assuming 4-bit priority registers, if PR1 initially contained the binary value “0001” where the ‘1’ was in the bit-0 position, data source/processing resource channel 1 would be of the highest priority. PR2-PR4 would contain “0010”, “0100”, and “1000” respectively, indicating data source/processing resource channels 2, 3 and 4 would have second, third, and fourth highest priorities respectively.
Likewise if PR1 initially contained the binary value “0100” where the ‘1’ was in the bit-2 position, data source/processing resource channel 3 would be of the highest priority. PR2-PR4 would contain “1000”, “0001”, and “0010” respectively, indicating data source/processing resource channels 4, 1 and 2 would have second, third, and fourth highest priorities respectively.
The RPQ makes arbitration decisions for directing input source data to the corresponding processing resources based on the contents of the priority registers PR1-PRn and status signals received by the RPQ indicating the availability of data from the source channels and the readiness of the processing resources. This is accomplished through the use of “ready” (RDY) signals issued by the data sources and “request” (REQ) signals issued by the processing resources. When a given data source channel has data to be processed, it issues a RDY signal to the RPQ. When a given processing resource channel is ready to accept new data, it issues a REQ signal to the RPQ.
If both a RDY and a REQ signal are pending for the channel specified by PR1 (i.e. the channel currently assigned as the highest priority), then the RPQ will set the n-bit SEL output signal such that the bit corresponding to the channel specified by PR1 is asserted and all other bits are zero. Additionally, the contents of the n priority registers will be “rotated” by one bit. Rotating consists of moving the asserted bit in a priority register to the next greater position of significance.
For example, assuming 4-bit priority registers, if PR1-PR4 contain the binary numbers “0100”, “1000”, “0001” and “0010” and RDY and REQ signals were pending for channel 3, SEL would be set as “0100.” The SEL signal would be routed to the switching circuitry which would allow data source/processing resource channel 3 to access the shared data path. Additionally, as the selected channel was associated with PR1, the priority registers would be rotated by one bit. Rotation by one bit would result in priority registers PR1-PR4 having new values “1000”, “0001”, “0010” and “0100” respectively. Therefore, data source/processing resource channel 4 would now be assigned the highest priority for the next round of arbitration.
Similarly, if either a RDY or a REQ signal are not pending for the channels specified by PR1-PRi-1 (i.e. the data source/processing resource channels currently assigned as the highest priority through the i−1th priority) but both a RDY and a REQ signal are pending for the data source/processing resource channel specified by PRi (i.e. the data source/processing resource channel currently assigned as the ith highest priority) then the RPQ will set the n-bit SEL output signal such that the switching circuitry would allow the data source/processing resource channel specified by the contents of PRi to access the shared data path. The contents of the n priority registers are also rotated by i bits.
For example, assuming 4-bit priority registers, if PR1-PR4 contain the binary numbers “0100”, “1000”, “0001” and “0010” and RDY and REQ signals were pending only for channel 1, the SEL signal would be routed to the switching circuitry which would allow data source/processing resource channel 1 access to the shared data path. Additionally, the priority registers would be rotated three bits such that PR1-PR4 would have new values “0010”, “0100”, “1000” and “0001” respectively. Therefore, data source/processing resource channel 2 would now be assigned the highest priority and most recently processed channel 1 would be assigned the lowest priority.
In another embodiment of the invention, the calculation of the SEL signal output and the rotation of the priority registers occurs in such a manner that, in the event that more than one data source/processing resource channel pair have RDY and REQ signals pending simultaneously, the maximum latency for processing the lowest priority channel is bounded, deterministic, and minimal. In another embodiment, this maximum latency is determined solely by the number of channels in the system and not the level of activity on the channels. In still another embodiment, the maximum latency is n clock cycles where n is the total number of channels.
In another embodiment of the invention, the design of the RPQ ensures that the channel which has just been designated for processing will have the lowest priority for the next round of arbitration. This results in an arbitration scheme having an inherent fairness where no single channel is given preference over another except as a result of the actual capabilities/needs of the channels.
In another embodiment of the invention, the design of the RPQ ensures that no channel may be prevented from being serviced (i.e. “starved”) by other channels, even one making high-rate requests (possibly every clock tick). In a system with n channels, the design bounds the potential wait for access to the interface at n clock ticks as all the other channels are serviced, at most, one time each.
In another embodiment of the invention, in the event that a channel pair is signaling both REQ and RDY, the channel must be serviced immediately if, in the current priority state defined by the priority registers, no higher priority channel is eligible for processing. In other words, the absence of pending REQ or RDY signals for higher priority channels does not prevent shared data path access by lower priority channel that has REQ and RDY signals pending.
It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the invention claimed. The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate an embodiment of the invention and together with the general description, serve to explain the principles of the invention.
The numerous objects and advantages of the present invention may be better understood by those skilled in the art by reference to the accompanying figures in which:
Reference will now be made, in detail, to presently preferred embodiments of the invention, an example of which is illustrated in the accompanying drawings.
Referring to
Referring to
Referring to
Processing resource channels may issue a request (REQ) signal 310-313 when they are prepared to accept input data. Data source channels may issue a ready (RDY) signal 320-323 when input data is valid and ready to be processed. The output of the RPQ is a four-bit vector SEL 330 containing at most one bit asserted. The SEL signal is routed to switching circuitry which regulates data source/processing resource channel access to the shared data path. The asserted bit in the SEL signal indicates which channel is permitted access (i.e. the assertion of bit-2 permits channel 3 to access the shared data path.)
Priority registers PR1-PR4 341-344 maintain binary values wherein only one bit is asserted at a given time. Each bit of the four-bit register corresponds to a data source/processing resource channel (i.e. bit zero to channel one, bit one to channel two, etc.) An example of this mapping scheme is depicted in
Priority register PR1 341 may initially contain the binary value “0001”, with the ‘1’ in bit-zero indicating that channel one currently is of the highest priority. PR2 342 may initially contain “0010”, with the ‘1’ in bit-one indicating channel two is of second highest priority. Similarly, registers PR3 343 and PR4 344 maintain values corresponding to the third and the fourth priority channels, respectively.
The output of the first AND gate 350 reports which channels have pending REQ and RDY signals. The outputs 361-364 of the column of four AND gates 360 fed by priority registers 341-344 indicate whether the channel which currently has the particular priority specified by a given priority register has both a pending RDY and REQ.
The three AND gates 370 (5, 6 and 7-bit AND gates respectively) and their resulting signals HIT2-HIT4 372-374 combined with the three OR gates 380 and their resulting signals ROT1-ROT3 381-383 (and their inverses !ROT1-!ROT3) define the four prioritization levels.
For example, if both a RDY and a REQ are pending for the channel specified by register PR1 341, (i.e. REQ1 310 and RDY1 320 are asserted and PR1=“0001”) then HIT1 361 will be nonzero, and ROT1 381 will be a ‘1’. In this case, the contents of all four priority registers will be rotated by one bit. If PR1 341 initially contains “0001”, the next value will be “0010”, indicating that channel two has become the highest priority channel. Similarly, PR2 342 initially contains “0010”, PR3 343 contains “0100”, and PR4 344 contains “1000”. The next values for PR2 through PR4 will be “0100”, “1000”, and “0001” respectively.
If either a RDY or REQ are not asserted for the highest priority channel as specified by PR1 341, but both are pending for the lower priority channel as (i.e. either REQ1 310 or RDY, 320 are not asserted, and REQ2 311 and RDY2 321 are asserted and PR2=“0010”), then HIT1 361 will be zero, ROT1 381 will be a ‘0’, HIT2 372 will be non-zero and ROT2 382 will be a ‘1’. In this case, the contents of the four priority registers will be rotated by two bits. If PR1 341 contains “0001”, the next state will be “0100” indicating that the highest priority channel for the next round of arbitration will be channel 3. Similarly, if PR2 342 contains “0010”, the next state will be “1000” indicating that the second highest priority for the next round of arbitration will be channel 4. The rotating process continues through the remaining channels and is easily extensible to more than four channels.
Referring to
Referring to
Input signals ROT1-ROT3 411-413 correspond to output signals ROT1-ROT3 381-383 of
For example, if the current state 440 of register 430 is the binary value “0100” and no rotate signal 411-413 is asserted, the enable signal 421 to the register 430 would be ‘0’ and the register would not load a value on the next clock cycle. Additionally, the switching circuit would maintain the default value and the next state 460 of register 430 would remain “0100.”
However, if the current state 440 of register 430 is the binary value “0100” and rotate signal ROT2 412 is asserted, the enable signal 421 to the register 430 would be ‘1’ and the register would load a new value on the next clock cycle. The next state 460 of register 430 will correspond to the result of the rotation calculation 442 (i.e. rotating two bits would result in the value “0001”) dictated by the asserted rotate signal ROT2 412.
Referring to
When a particular channel has both a RDY and a REQ pending, that channel is enqueued for processing. This channel number is compared to the value currently stored in Priority Register 1 (PR1) 550. If the subject channel number is the same as that stored in PR1, the bit corresponding to the subject channel number of the n-bit “select” (SEL) output signal is set as ‘1’, all other bits are set as ‘0’, and the contents of the priority registers PR1 through PR4 are rotated one bit 570.
If the subject channel number is not the same as that stored in PR1, the channel number is compared to the successive priority registers PR2-PRi-PRn until a match is found 560. When a match is found, the bit of the n-bit SEL output signal corresponding to the channel number encoded in the PRi register is set as ‘1’, all other bits are set as ‘0’, and the contents of all PR registers are rotated i bits 580
Once the SEL bit has been set and the priority registers have been appropriately rotated, the RDY and REQ signals for the processed channel are deasserted 590 and the system waits for another pair of RDY and REQ signals to be asserted for a given channel.
It is believed that the present invention and many of its attendant advantages will be understood from the foregoing description, and it will be apparent that various changes may be made in the form, construction, and arrangement of the components thereof without departing from the scope and spirit of the invention or without sacrificing all of its material advantages. The form herein before described being merely an explanatory embodiment thereof, it is the intention of the following claims to encompass and include such changes.
Bortz, Mark A., Hiratzka, T. Douglas, Limondin, Philippe M.
| Patent | Priority | Assignee | Title |
| 8127063, | Jan 20 2009 | Fisher-Rosemount Systems, Inc. | Distributed equipment arbitration in a process control system |
| 8612649, | Dec 17 2010 | AT&T Intellectual Property I, L P | Validation of priority queue processing |
| 8744347, | Aug 04 2006 | VERIMATRIX | Method of routing incoming application data in an NFC chipset, for identification of the application |
| Patent | Priority | Assignee | Title |
| 4924380, | Jun 20 1988 | MODULAR COMPUTER SYSTEMS, INC A FLORIDA CORPORATION | Dual rotating priority arbitration method for a multiprocessor memory bus |
| 5564062, | Mar 31 1995 | International Business Machines Corporation | Resource arbitration system with resource checking and lockout avoidance |
| 6160812, | May 04 1998 | WSOU Investments, LLC | Method and apparatus for supplying requests to a scheduler in an input buffered multiport switch |
| 7143219, | Dec 31 2002 | Intel Corporation | Multilevel fair priority round robin arbiter |
| 20060277330, |
| Executed on | Assignor | Assignee | Conveyance | Frame | Reel | Doc |
| Aug 23 2006 | Rockwell Collins, Inc. | (assignment on the face of the patent) | / | |||
| Aug 23 2006 | HIRATZKA, T DOUGLAS | Rockwell Collins, Inc | ASSIGNMENT OF ASSIGNORS INTEREST SEE DOCUMENT FOR DETAILS | 018228 | /0667 | |
| Aug 23 2006 | LIMONDIN, PHILIPPE M | Rockwell Collins, Inc | ASSIGNMENT OF ASSIGNORS INTEREST SEE DOCUMENT FOR DETAILS | 018228 | /0667 | |
| Aug 23 2006 | BORTZ, MARK A | Rockwell Collins, Inc | ASSIGNMENT OF ASSIGNORS INTEREST SEE DOCUMENT FOR DETAILS | 018228 | /0667 |
| Date | Maintenance Fee Events |
| May 11 2012 | M1551: Payment of Maintenance Fee, 4th Year, Large Entity. |
| May 11 2016 | M1552: Payment of Maintenance Fee, 8th Year, Large Entity. |
| Apr 22 2020 | M1553: Payment of Maintenance Fee, 12th Year, Large Entity. |
| Date | Maintenance Schedule |
| Nov 11 2011 | 4 years fee payment window open |
| May 11 2012 | 6 months grace period start (w surcharge) |
| Nov 11 2012 | patent expiry (for year 4) |
| Nov 11 2014 | 2 years to revive unintentionally abandoned end. (for year 4) |
| Nov 11 2015 | 8 years fee payment window open |
| May 11 2016 | 6 months grace period start (w surcharge) |
| Nov 11 2016 | patent expiry (for year 8) |
| Nov 11 2018 | 2 years to revive unintentionally abandoned end. (for year 8) |
| Nov 11 2019 | 12 years fee payment window open |
| May 11 2020 | 6 months grace period start (w surcharge) |
| Nov 11 2020 | patent expiry (for year 12) |
| Nov 11 2022 | 2 years to revive unintentionally abandoned end. (for year 12) |